TerryS;675613 Wrote: > Gee, Sorry I caused you to waste your time yet again. I'm not being > stubborn (all stubborn people say that). I'm truly looking for > information. I am an Electrical Engineer. I have been designing ADC > and DAC circuits for years. I know about dithering. I use a spectrum > analyzer almost daily. I also use oscilloscopes in FFT mode to view > the spectrum of signals. I often have to pull signals out of the noise > floor to measure them. I even manually calculated FFTs on time domain > signals in school (just to prove it can be done I guess). > ....... > > Dithering on a sine wave, does de-correlate the quantization error > (distortion) from the sinal, and through noise shaping moves it out of > the audible range. It doesn't prevent the distortion. It just moves > the harmonic components out of band. Is this the same as having a > distortion free signal, or just a trick to make the distortion sound > less objectionable? > > We all 'know' that dithering can effectively increase the resolution of > a ADC or DAC. But how much? Have you ever come across a number for > this? Why not? Because it depends.... It depends on how many samples > you can make of the signal before it changes. The more samples I can > make on the waveform, the better it works. One sample does nothing. > Thousands of samples will pull a signal out of the noise floor like > magic. But I don't see how I can expect many samples on a music > waveform. > If dithering worked so well, why do I even need 16 bits? Wouldn't 8 > bits work with enough dithering? Why not just one bit? Well like you > pointed out, I could just use one bit. I just have to make the > sampling rate really, really high. > Show me the analysis that shows how much improvement dithering gets me > on a typical music waveform sampled at 44kHz. I've been looking for 20 > years.... > > Terry There is an interesting point buried in here, but there are some really bad ones mixed up. There isn;t much analysis i can find about for example real world musical transients. I have fouind this very annoying because i have been trying to get to the bottom of the question of whether the impulse response of a perfect 22kHz ish brick wall linear phase filter would really call any perceptible time domain smearing on a real world trasnient recording.
I agree that generally the examples used employ tones not real world signals, but I don;t really see that that makes much difference in this case. I can;t show you the analysis you seek, but then again I can;t see any analysuis of the problem of undithered 16 bit data either in order to make the comparison. Ultimately the point is that the resolution is basically set by the limits of human hearing. Essentially the problem just comes back to the question of resolution of information in the presence of noise. A dithered signal is just the same as an analog signal plus noise. The question of resolution of signal below the noise is the same whether you are asking about the analog signal plus noise or the dithered digital system I will not bother to go onto your "music continuously changing" point save to say that I think you are confusing the change in sample values over time, with the change of frequency components represnted by those sample values in the discrete time domain. a) if you are as you claim and Electical Engineer that why won't you undertstand the difference betrween noise and distortion? Whjy cant you tell the difference in FFT terms between a signal plus noise and a signal plus distortion? That is all you need to know to understand the effect of dither. This talk of "It doesn't prevent the distortion. It just moves the harmonic components out of band." ....is just plain WRONG: there are no harmonic components of the quantisation error after it has been dithered. You do not have to analyse the result over 32000 samples to grasp this. b) If have read the link from Jim Lesurf which I posted ages ago you would have read that the use of a 0.7s interval for FFTs is conventional, but the same principle would apply with other intervals. The question of which interval would produce the most meaningful result somewhat moot but I think it should be considered in terms of pyschoacoustics as the ultimate question is over what period does human hearing gernate its "picture" of sound. It is my (very imperfect) understadnign that human hearing does in fact work like FFT and that there is a limit to the time as well as frequecny resolution. I do not know whether the .7s interval is psychoacoustically significant, but it may be I understand that a tone has to continue for a certin time befroe it will be perceived as such by a human being. perhaps someone can help with the precise amount but i believe it will be several mS. Over that period there will be 44x as many samples as the duration in ms. I may have got the maths wrong but I reckon that over 6 ms ie 254 samples(which I think is the minumum time over which two distinct sounds will be heard as separate souinds) the noise floor for properly dithered 16 bit signals would appear as about -115 dB rather -130 dB as it usually appears in the 0.7s FFT.(I think this would be about 24 dB below the dithered noise power of -93dB) This would be ample to show that a tone well under -93dB can be resolved without noise shaping. This lowering of the noise floor on the FFT is not the effect of the dither but the effect of the interval over which the FFT is taken and the number of discrete frequency bins it looks at. If you want a better answer on this why not ask at hydrogenaudio where there are lots of people who really understand this stuff. -- adamdea ------------------------------------------------------------------------ adamdea's Profile: http://forums.slimdevices.com/member.php?userid=37603 View this thread: http://forums.slimdevices.com/showthread.php?t=89733 _______________________________________________ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/audiophiles
